Differentiation in ABA: Reading a Clear Effect

Differentiation means the data show a clear difference between conditions. You look at a graph and see a real gap. One condition sits high. Another sits low. That visible separation is differentiation.

This idea matters most when you read graphs. BCBAs use it to judge if an effect is real. It helps in functional analysis and in treatment. Without differentiation, you cannot trust your conclusion.

What differentiation looks like#

Differentiation is about a clear, visible difference. You are not running a statistics test. You are looking at the pattern with trained eyes. Matt Harrington describes the moment you see it.

Differentiation is when you and I, as the behavior analyst, can look at that graph and say, yep, there's a clear difference, there's a clear difference on the dependent variable based on the independent variables that we put into play. From the talk — Matt Harrington

Let us unpack the terms. The dependent variable is the behavior you measure. The independent variables are the conditions you set up. Differentiation means the behavior clearly changed with the conditions.

So the graph does the talking. If two conditions overlap, you have poor differentiation. If they pull apart cleanly, you have strong differentiation. The gap is your evidence.

Why differentiation matters#

Differentiation is the proof that your conditions did something. A flat, overlapping graph tells you little. You cannot say the conditions caused a change. You are left guessing.

A clearly separated graph tells a clear story. It shows the behavior tracked your conditions. That gives you the confidence to act. You can pick a function or a treatment with reason behind it.

This protects your clients, too. Weak evidence can lead to a weak plan. A poor plan wastes time and may not help. Strong differentiation keeps your decisions grounded in data.

The standard behind a "clear" effect#

"Clear" can sound like an opinion. But there is a standard behind it. Harrington points to the What Works Clearinghouse. That group sets rules for single-case design research.

Well, pulling from the What Works Clearinghouse, which is kind of an organization that sets standards for single-case design research, as well as meta-analyses and things like that, what they're looking for is, in quote, at least three demonstrations of an intervention effect at at least three different points in times with reasonable certainty that the observed data are sufficient to capture important information about the patterns of responding. From the talk — Matt Harrington

Notice the key number. They want at least three demonstrations of an effect. They want those at three different points in time. One good-looking spike is not enough.

This raises the bar in a helpful way. You are not fooled by a single lucky data point. You look for a pattern that repeats. Repetition is what makes an effect believable.

How this plays out in a functional analysis#

A functional analysis tests why a behavior happens. You run different conditions on purpose. Each condition targets a possible reason for the behavior. Then you read the graph for differentiation.

Say one condition shows high behavior every time. The other conditions stay low. That clear split points to the function. The behavior lines up with one clear cause.

But sometimes the graph does not separate. The conditions overlap and jump around. That is undifferentiated data. It means you cannot yet name the function with confidence.

When that happens, you do not force an answer. You go back and adjust. Maybe your conditions were not distinct enough. Maybe you need more sessions to see a stable pattern.

What weak differentiation is telling you#

Undifferentiated data is not a dead end. It is useful information. It says your current setup did not reveal a clear effect. That is a signal to change something.

Common causes are worth checking. Your conditions may be too similar to each other. Your sessions may be too short to show a pattern. Outside factors may be adding noise to the data.

Fixing these often sharpens the picture. You make the conditions more distinct. You run enough sessions to see a stable trend. Then you look again for that clean separation.

How to strengthen your read#

Start with clear, distinct conditions. If conditions blur together, the graph will blur too. Design each one to test a single idea. That gives the data room to separate.

Next, collect enough data across time. Keep the three-demonstration standard in mind. One session cannot prove much on its own. A repeated pattern carries far more weight.

Finally, be honest about what you see. If the graph does not separate, say so. Do not read a clear effect into a messy graph. Waiting for real differentiation is the stronger move.

What to look at on the graph#

Differentiation is a skill you can train your eyes for. A few features help you judge it. You look at level, which is how high or low the data sit. You look at trend, which is the direction over time.

You also watch how much the data bounce around. High bounce, called variability, can hide a real effect. Low bounce makes a gap easier to trust. Steady data with a clear split is the goal.

Overlap is another key feature. Ask how much the conditions share the same range. Little overlap points to strong differentiation. Heavy overlap points to weak differentiation.

Put these together and the picture gets clear. Separate levels, low overlap, and stable data signal a real effect. Mixed levels and heavy overlap signal caution. Reading these features keeps your judgment honest.

FAQ#

What does differentiation mean in ABA?

Differentiation means the data show a clear difference between conditions. You look at a graph and see a real gap in the behavior. It tells you the behavior changed along with your conditions. It is the visible proof that an effect is real.

How much difference counts as differentiation?

There is no single cutoff, but the pattern must repeat. A common standard asks for at least three demonstrations of an effect at three different times. One spike is not enough. You want a clear, repeating separation you can see on the graph.

What should I do if my data are not differentiated?

Treat it as useful feedback, not failure. Undifferentiated data often means the conditions were too similar or the sessions too few. Make your conditions more distinct and collect more data across time. Then check the graph again for a clean gap.

You can watch this skill taught step by step in Confessions of a New Behavior Analyst in Functional Analysis.